OpenCode launched Go, a $10/month subscription that bundles 12 open-source coding models behind a single API key and OpenAI-compatible endpoint. The service works with OpenCode’s own CLI, but the compatibility layer means it drops into any tool that accepts OpenAI-format API calls, including OpenClaw agents, Hermes Agent, and Pi Agent, according to a hands-on review by BitDoze.
The model roster includes DeepSeek V4 Pro, DeepSeek V4 Flash, GLM 5.1, GLM 5, Qwen 3.6 Plus, Qwen 3.5 Plus, Kimi K2.5, Kimi K2.6, MiniMax M2.7, MiniMax M2.5, MiMo V2.5 Pro, and MiMo V2.5. Models are hosted across the US, EU, and Singapore with a zero-retention policy, meaning code sent through the API is not used for provider training.
Pricing and Limits
Go uses dollar-value caps rather than request counts. The monthly limit is $60 in model value, with a $12 cap per five-hour window and $30 per week. Actual request counts vary by model. DeepSeek V4 Flash, the cheapest option, allows roughly 158,000 requests per month. GLM 5.1, the most expensive, caps at around 4,300, per BitDoze’s breakdown.
Users who need burst capacity beyond Go limits can enable fallback to their OpenCode Zen balance, which charges per-token at standard rates.
The Builder Economics Angle
The value proposition targets indie developers and small teams running coding agents who want to avoid managing multiple API keys across providers. At $10/month for 12 models, Go undercuts the per-token cost of running these same models individually through OpenRouter or direct provider APIs for most usage patterns.
The OpenAI-compatible endpoint is the critical feature. Builders running OpenClaw agents or Hermes sessions can swap in Go by changing a base URL and API key, with no code changes required. The base URL is https://opencode.ai/zen/go/v1/chat/completions.
The OpenCode team curates the model list rather than offering every available open-source model, testing each for coding agent performance and working with model teams to optimize serving, according to BitDoze.